[PATCH] D76365: [cuda][hip] Add CUDA builtin surface/texture reference support.

[Michael Liao via Phabricator via cfe-commits]

hliao marked 6 inline comments as done.
hliao added inline comments.


================
Comment at: clang/lib/CodeGen/CodeGenModule.cpp:701-713
+  if (getLangOpts().CUDAIsDevice) {
+    // As CUDA builtin surface/texture types are replaced, skip generating TBAA
+    // access info.
+    if (AccessType->isCUDADeviceBuiltinSurfaceType()) {
+      if (getTargetCodeGenInfo().getCUDADeviceBuiltinSurfaceDeviceType() !=
+          nullptr)
+        return TBAAAccessInfo();
----------------
tra wrote:
> hliao wrote:
> > tra wrote:
> > > Would `isCUDADeviceBuiltinTextureType()` be sufficient criteria for skipping TBAA regeneration?
> > > Or does it need to be 'it is the texture type and it will be replaced with something else'? What is 'something else' is the same type?
> > > 
> > > 
> > The replacement only happens in the device compilation. On the host-side, the original type is still used.
> But you've already checked CUDAIsDevice so you already know that you want to replace the type.
> `if (getTargetCodeGenInfo().getCUDADeviceBuiltinTextureDeviceType() !=  nullptr)` appears to be redundant and can probably be dropped.
That check is a target-specific one, which may choose very different implementation on how to handle these builtin surface/texture types. If they don't want to change those types on the device side and, instead, use very different different `textureReference`. Their `getCUDADeviceBuiltinTextureDeviceType()` may return `nullptr` to keep use the same reference type on both host- and device-side compilation.


================
Comment at: clang/lib/CodeGen/CodeGenModule.cpp:4101-4127
+        if (const ClassTemplateSpecializationDecl *TD =
+                dyn_cast<ClassTemplateSpecializationDecl>(RD)) {
+          Linkage = llvm::GlobalValue::InternalLinkage;
+          const TemplateArgumentList &Args = TD->getTemplateInstantiationArgs();
+          if (RD->hasAttr<CUDADeviceBuiltinSurfaceTypeAttr>()) {
+            assert(Args.size() == 2 &&
+                   "Unexpcted number of template arguments of CUDA device "
----------------
tra wrote:
> hliao wrote:
> > tra wrote:
> > > This is the part I'm not comfortable with.
> > > It's possible for the user to use the attribute on other types that do not match the expectations encoded here.
> > > We should not be failing with an assert here because that's *user* error, not a compiler bug.
> > > 
> > > Expectations we have for the types should be enforced by Sema and compiler should produce proper diagnostics.
> > > 
> > `device_builtin_surface_type` and `device_builtin_texture_type` should only be used internally. Regular users of either CUDA or HIP must not use them as they need special internal handling and coordination beyond the compiler itself.
> I agree that it's probably not something that should be used by users.
> Still, such use should be reported as an error and should *not* crash the compiler. Asserts are for clang/llvm developers to catch the bugs in the compiler itself, not for the end users misusing something they should not. 
> 
addressed in the latest revision


================
Comment at: clang/lib/CodeGen/TargetInfo.cpp:6471-6472
+    // Lookup `addrspacecast` through the constant pointer if any.
+    if (auto *ASC = llvm::dyn_cast_or_null<llvm::AddrSpaceCastOperator>(C))
+      C = llvm::cast<llvm::Constant>(ASC->getPointerOperand());
+    if (auto *GV = llvm::dyn_cast_or_null<llvm::GlobalVariable>(C)) {
----------------
tra wrote:
> hliao wrote:
> > tra wrote:
> > > What's the expectation here? Do we care which address spaces we're casting to/from? 
> > We need to check whether we copy from that global variable directly. As all pointers are generic ones, the code here is to look through the `addrspacecast` constant expression for the original global variable.
> I'm still not sure what exactly you want to do here.
> If the assumption is that all `addrspacecast` ops you may see are from global to generic AS, this assumption is not always valid. I can [[ https://clang.llvm.org/docs/LanguageExtensions.html#memory-references-to-specified-segments | annotate any pointer with an arbitrary address space ]] which may then be cast to generic. Or something else.
> 
> If you accept Src as is, without special-casing addrspacecast, what's going to happen?
> AFAICT `nvvm_texsurf_handle_internal` does not really care about specific AS.
the backend needs a GlobalVariable as the argument for that intrinsic. The lookup through `addrspacecast` to check a global variable, which is created in the global address space and casted into a generic pointer.


================
Comment at: clang/lib/Headers/__clang_cuda_runtime_wrapper.h:82-94
 #undef __CUDACC__
 #if CUDA_VERSION < 9000
 #define __CUDABE__
 #else
+#define __CUDACC__
 #define __CUDA_LIBDEVICE__
 #endif
----------------
tra wrote:
> hliao wrote:
> > tra wrote:
> > > Please add comments on why __CUDACC__ is needed for driver_types.h here? AFAICT, driver_types.h does not have any conditionals that depend on __CUDACC__. What happens if it's not defined.
> > > 
> > > 
> > `driver_types.h` includes `host_defines.h`, where macros `__device_builtin_surface_type__` and `__device_builtin_texture_type__` are conditional defined if `__CUDACC__`.
> > 
> > The following is extracted from `cuda/crt/host_defines.h`
> > 
> > ```
> > #if !defined(__CUDACC__)
> > #define __device_builtin__
> > #define __device_builtin_texture_type__
> > #define __device_builtin_surface_type__
> > #define __cudart_builtin__
> > #else /* defined(__CUDACC__) */
> > #define __device_builtin__ \
> >         __location__(device_builtin)
> > #define __device_builtin_texture_type__ \
> >         __location__(device_builtin_texture_type)
> > #define __device_builtin_surface_type__ \
> >         __location__(device_builtin_surface_type)
> > #define __cudart_builtin__ \
> >         __location__(cudart_builtin)
> > #endif /* !defined(__CUDACC__) */
> > ```
> My concern is -- what else is going to get defined? There are ~60 references to __CUDACC__ in CUDA-10.1 headers. The wrappers are fragile enough that there's a good chance something may break. It does not help that my CUDA build bot decided to die just after we switched to work-from-home, so there will be no early warning if something goes wrong.
> 
> If all we need are the macros above, we may just define them. 
Let me check all CUDA SDK through their dockers. Redefining sounds good me as wll.


================
Comment at: clang/test/CodeGenCUDA/surface.cu:12-14
+template<typename T, int type = 1>
+struct __attribute__((device_builtin_surface_type)) surface : public surfaceReference {
+};
----------------
tra wrote:
> Please add a test for applying the attribute to a wrong type. I.e. a non-template or a template with different number or kinds of parameters. We should have a proper syntax error and not a compiler crash or silent failure.
addressed in refined tests in the latest revision


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D76365/new/

https://reviews.llvm.org/D76365